一种用于人脸识别的基于主导近邻像素的局部Gabor空间直方图特征  

Feature of Local Gabor Spatial Histogram based on Dominant Neighboring Pixel for Face Recognition

作　　者：刘军[1] 景晓军[1] 孙松林[1] 连自锋[1] 

机构地区：[1]北京邮电大学信息与通信工程学院,北京100876

出　　处：《北京邮电大学学报》2015年第1期51-54,共4页Journal of Beijing University of Posts and Telecommunications

基　　金：国家自然科学基金项目(61143008)

摘　　要：对于实时性要求较高的人脸识别算法,图像特征表示不仅应该具有高的区分能力,还应该具有较小的空间维数.提出了一种基于主导近邻像素的局部Gabor空间直方图(LGSH-DNP)的图像特征表示方法.首先,对人脸图像进行Gabor滤波,滤波后图像上的每个像素用其邻近的两个最大的像素的位置表示并编码成该像素的描述符;然后,提取描述符所形成图像的空间直方图;最后,采用加权交集直方图相似度来度量空间直方图之间的距离,从而实现人脸图像识别.在人脸识别技术图像库上的实验结果表明,该图像特征算法具有较高的识别率和较低的计算复杂度.For face recognition system used in real-time application,the image feature representation should not only have high discriminative capability,but also the low dimensionality. A new face image representation-local Gabor spatial histogram based on the dominant neighboring pixel（ LGSH-DNP） was proposed. Firstly,the face image is filtered by the Gabor filter bank. Each pixel of the resulted filtered images is labeled by the positions of the two neighboring pixels with the highest values,resulting in multiple descriptor images. Secondly,the spatial histogram is extracted from these descriptor images. Finally the weighted intersection histogram similarity measure is used to realize face classification. Extensive experiments are performed on face recognition technology face image database. It verifies the effectiveness of the proposed LGSH-DNP method.

关 键 词：局部Gabor空间直方图 主导近邻像素 Gabor滤波响应幅度图像 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]